Add geolocation blog dataset

blog_data/by_source/benjaminstrick.jsonl

@@ -0,0 +1,4 @@
+{"source":"benjaminstrick","url":"https://benjaminstrick.com/boko-haram-nigeria","title":"Boko Haram\\u2019s systematic destruction and displacement of Nigerian communities","status":"ok","retrieved_at":"2026-08-29T16:45:46.337602+08:00","text":"Boko Haram’s systematic destruction and displacement of Nigerian communities · Benjamin Strick BS Benjamin Strick Open-source investigations All writing Work ← Back to site Satellite Case study Boko Haram’s systematic destruction and displacement of Nigerian communities Published 8 August 2018 Read 9 min Villages, towns and dense areas have been burnt, destroyed, and completely levelled in northern Nigeria as the country continues its fight against the Boko Haram insurgency. Using Google Earth’s history function, we can view how landscapes change over years, as well as study developments in infrastructure and refugee movement through conflict areas. For the northern region of Nigeria, the landscape tells a grim tale. In Nigeria’s Borno State, what was once a thriving and culturally rich area, is now a void land as the population shelters in siege cities where the frontlines are protected by trenches, literally. Villages now have only their foundations remaining after they were burnt to the ground. The clear path of destruction mapped out by Boko Haram’s systematic approach to village destruction in Borno State from 2014 to 2017 can be seen as they jump from town to town destroying everything they leave behind. One roadside village was functioning in February 2016, but by March 2017 only foundations and ash remain. Another scene is the village of Mutube with its surrounding communities: the satellite image from 2013 shows a village, and the next available image from 2017 shows only trees remaining. For many communities that have been destroyed, there is no hope of rebuilding. That being said, the Nigerian Government is urging some refugees to return home . But there is nothing left, and if there was, it would again be under attack by Boko Haram insurgents still at large in the area. More recent village destruction can be seen below, where only a few foundations of buildings remain in the space of 11 months since 2017. Eleven months apart Mapping the disappearance of hundreds of sub-communities Surrounding the Sambisa Forest in Nigeria’s north is where the majority of the evidence of displacement can be viewed. According to UNHCR , that count of displacement is in the millions. The forest itself covers a very large and difficult terrain where many of Boko Haram’s attacks have been launched from. On the outskirts of the forest, traditional villages are all too common a story. When we view their status in the available 2017 map of Google Earth, there are no visible signs of what once stood there. There are hundreds of villages and towns like this that have been erased off the map. Three such communities were destroyed at some point between March and November 2016. Significantly larger and denser than the small villages, the town of Nguro-Soye was the victim of many suicide bombings and attacks by Boko Haram militants. Attacks on crucial infrastructure such as a telecommunications tower and suicide bombings of the town’s marketplace led to a downfall of Nguro-Soye, and its ultimate destruction. The satellite image from March 2018 shows few remnants after the town’s destruction. Further south of Nguro-Soye, on a T-intersection of a main road, we are able to see more defined darker areas where villages once stood, indicative of burning. These villages would have also been prime targets given the accessibility of the main highway for Boko Haram and Nigerian military to operate in the area. In a bid to control this area, the Nigerian military have used their \"trench-laying\" tactic to fortify the position against suicide bombers. Many of these trenches are scaled at 2 to 4m wide and 2 to 3m deep, and are a quick and effective solution to protect areas against potential suicide bombers or vehicle attacks, two offensive tactics Boko Haram employ. For villages that have been built along main roads, in a more peaceful time it would have been a good thing. But for villages in northern Nigeria, it means they are more accessible to Boko Haram. Almost all villages constructed along main roads near the major fortified town of Bama were systematically wiped out between 2016 and 2017. This is the case for the small town of Mairamri, 15km south-east of Bama. More recent destruction occurred at the village of Gajibo in Borno State along the Gamboru Ngala Highway. News reports from July 2018 state 27 were killed in an attack on the village, however, we can see from satellite images that the town has been largely destroyed, bar several shelters. Gajibo, before and after Any villages that are still standing are in the process of being besieged by Boko Haram, or are strongholds of Boko Haram, and face attacks and bombing by Nigerian military. Smoke signals from bombings and attacks are a common story. Dikwa: the siege city offering refuge With so much displacement from homes, communities and towns being destroyed, where have the majority of refugees gone? There is evidence in a number of larger towns that have the afforded protection of the siege trench walls and strong military presence. One such area is the town of Dikwa . Dikwa’s history has seen the town pivoted as a stronghold since the 1800s, and has seen German and French occupancy. Now the town is a well-protected sanctuary of the Borno State. It uses the first line of defence trench system encircling the town. The only access points in the town are main roads, which are heavily guarded with stringent checkpoints, visible in the satellite image from 2018. Due to the safety Dikwa is able to provide, it has been established as a refugee-haven for many of the displaced villagers from the outlying areas. We can see this in evidence of the rapidly erected refugee cities in Dikwa. The living conditions in these camps are dire. The capital of Borno State, and the area’s largest city, Maiduguri , is also a haven for many of the displaced with its well-protected boundaries. Since village destruction began in 2015, Maiduguri has struggled to house refugees. In the space of three years, a refugee city is born. The millions of displaced Nigerians, referred to as IDPs or Internally Displaced People, only need to look past city defences to see the plight of villages that sit beyond its boundaries. Beyond the protection of Dikwa’s boundary, one small village was levelled between 2016 and 2017. Outside the Dikwa perimeter, 2016 and 2017 Bitta: the failed siege town For the most part, the Nigerian defensive trench is able to protect the towns it surrounds. For Bitta , it was not able to do so. Bitta. Left: pre-2014. Right: December 2017 Bitta has been a hotspot for Boko Haram and Nigerian military clashes and many of its citizens have left after its destruction. The military has used Dikwa as a launchpad for the Borno State Nigerian military Deep Punch II Operation, focussing on the clearance of Boko Haram presence in the Sambisa Forest. The scars of clashes can be seen just beyond the guard boxes built along the north edge of the town. The refugees from villages like Bitta and others are not just moving to Nigeria’s refugee camps, but are also fleeing the danger and seeking refuge in bordering countries such as Chad and Cameroon. One of the largest external camps for Nigerian refugees is Minawao Refugee Camp in the Mayo Tsanaga region of Cameroon’s Far North. Minawao is exceeding numbers every day. The UN says Cameroon’s Far North region hosts more than 95,000 Nigerian refugees. Many of which are sheltering in Minawao. Since 2014, Minawao has become a city in Cameroon. Minawao, Cameroon. A barren landscape becomes a city While it did previously pledge a commitment to refugees, the Cameroonian Government has been \"forcibly returning\" a number of Nigerians. UNHCR say this is a clear violation of national and international laws. The fight against Boko Haram in 2018 still ensues, as regular attacks are reported against villages in Nigeria’s north, as well as bordering villages in Cameroon. Method This open-source case study was made using past and more present satellite imagery from Google Earth. I urge you to visit the communities via the links provided to see the destruction for yourself. Related writing All writing → Geolocation Adding geodata to Google Earth Pro → Geolocation Finding McAfee: geolocation and imagery analysis → GEOINT When the lights go out, satellites are watching → Get methods like this once a month OSINT Field Notes: tools, techniques and one case file. New editions are always free. Subscribe → © 2012 to 2026 Benjamin Strick benjaminstrick.com","image_urls":["https://benjaminstrick.com/images/blog/1-09ud44ncjaomdofkab1xkg-1341.png","https://benjaminstrick.com/images/blog/1-25tgsf7dh2pnbivbyzewxg-1024.png","https://benjaminstrick.com/images/blog/1-5qannf1hywgu-sf33-icag.png","https://benjaminstrick.com/images/blog/1-97vk4h3kl1xoirnnfsab4w-1024.png","https://benjaminstrick.com/images/blog/1-buv0rqe-nqjsgypq1sxslq.png","https://benjaminstrick.com/images/blog/1-ervuzn2dvravli5-urwp8a.png","https://benjaminstrick.com/images/blog/1-rlwagpmjweytveoueye3-q-1024.png","https://benjaminstrick.com/images/blog/1-uj-ep8y1nymf1ubfxoatbg.png","https://benjaminstrick.com/images/blog/1-x-fpu9myooprrg2tnyliyg-1024.png","https://benjaminstrick.com/images/blog/1-yijzcknn0vifcxigdqxtnq.png","https://benjaminstrick.com/images/blog/1-yzn95oyggdzjihjjpaxuwg.png"],"local_images":["blog_data/images/blog_517f0e7f0bb1_0.bin","blog_data/images/blog_517f0e7f0bb1_1.bin","blog_data/images/blog_517f0e7f0bb1_2.bin","blog_data/images/blog_517f0e7f0bb1_3.bin","blog_data/images/blog_517f0e7f0bb1_4.bin","blog_data/images/blog_517f0e7f0bb1_5.bin","blog_data/images/blog_517f0e7f0bb1_6.bin","blog_data/images/blog_517f0e7f0bb1_7.bin","blog_data/images/blog_517f0e7f0bb1_8.bin","blog_data/images/blog_517f0e7f0bb1_9.bin","blog_data/images/blog_517f0e7f0bb1_10.bin"]}
+{"source":"benjaminstrick","url":"https://benjaminstrick.com/finding-mcafee","title":"Finding McAfee: geolocation and imagery analysis","status":"ok","retrieved_at":"2026-08-29T16:45:39.335336+08:00","text":"Finding McAfee: geolocation and imagery analysis · Benjamin Strick BS Benjamin Strick Open-source investigations All writing Work ← Back to site Geolocation Case study Finding McAfee: geolocation and imagery analysis Published 15 March 2020 Read 6 min Identifying past, current, and possible future locations through the geolocation and chronolocation of media provided by a specific user. This case study is based on a challenge from well-known entrepreneur, John McAfee, to show how relative geolocation of two points on a chronological timeline can give a likely path and possible locations in between. To do this, two geographical points will be used, categorised by day, to geolocate a photo that was taken between those two points. This research post is split into the following four sections, in this order: Point B (the photo to geolocate) Point C (where the subject is traveling to) Point A (where and when the subject started the journey) Geolocation analysis of the Point B image The tools in this case study are completely free, so please do use them to follow along. I used Google Maps, GIMP (image editing) and Twitter. Contents 01 Point B: the image to geolocate 02 Point C: where the subject is traveling to 03 Point A: where and when the subject started the journey 04 Geolocation analysis of the Point B image Point B: the image to geolocate Below is the tweet in question. The challenge: self-explanatory. The tweet in question So where do you start in a case like this? Using a first approach to imagery intelligence (IMINT), look in the image and ask: \"what do I see?\". In the image, we have a number of clues that may indicate where this photo was taken. I am going to number some of them in the image below. Eight reference points to work from What do we see here? First, we have got McAfee. Who through his accounts may help give more clues. I will get to him later. This roof colouring gives an indication that it might be a brand colour. Coloured fuel bowsers indicate that this is both a fuel station, and is a unique identifier for what the brand or name of the fuel station may be. Large trucks use this fuel station, so it is likely that it is in an open area, or along a highway. There are flowers in front of the building McAfee is next to, which means it might be a store for the fuel station. There is a light blue band along the horizon. This is usually synonymous with a large body of water. This post and building would be a unique identifier on satellite imagery. The horizon is not cluttered with buildings or trees, indicating it might be a flat plain and out of built-up urban areas. That is a lot of reference points we have to go off. And now, since we have done an initial imagery analysis checklist, we can work our way down that list to investigate each of those leads. First is John McAfee. He is likely to indicate relevant information in his social media. Point C: where the subject is traveling to In the initial tweet above the subject indicated he was \"on the way to London\" and that the photo was taken in the past. This identifies our destination. How is he traveling there? Considering the location of the photo as we analysed in the eight takeaways above, it is clear he is at a service station with his large security crew. So it is likely he is driving to his destination. Point A: where and when the subject started the journey Where did the subject come from? This is where we can start using the intelligence tool I would like to refer to as \"geoprofiling\". Essentially we are going to map out a short chronological timeline of where McAfee was in order to find where he is. Scrolling back through his Twitter timeline, we can find this tweet. It was posted two days before the other photo. Two days earlier What is important about this tweet is it gives us a location as well as a destination. Take a look in the red box in the right of the image below. A signboard reading Hotel Schlicker It is in Munich, Germany . We know this is the place where the photo was taken as there are a number of features that match those seen in geotagged images on Google Maps and Facebook. First is the identical match of the sign and wall lining to this one found on Google Maps . Matching the sign, wall lining and ground paint against two independent sources Second, for further verification, we can identify both the sign and the white paint on the ground from this image on the Hotel Schlicker Facebook page. For chronolocation, for the purposes of this post there is no necessity to conduct a shadow calculation. Merely being able to plot this geolocated image on our timeline gives us an original lead for the following reference points: Hotel Schlicker in Munich, Germany on January 26, 2020 The unknown point in between London, at a future point in time Why is the second point unknown? Because that is the initial location we need to find. And we are going to do that now. Geolocation analysis of the Point B image We already know two things about the image. It was taken on the road between Hotel Schlicker and London. Using Google Maps \"direction\" feature we can make a simple indication of the route between those two points. A lot of empty space to cover That is a lot of empty space to cover. But we have a list of eight points we can use to filter that information down. First, what is the place they stopped at in the image? The features of the roof and fuel pumps will help with that. Those features are something we can use in a Google Search term. For this, I used what we know and what I see in the following string search for Google Images: It is likely in Europe, so that is a term I used: \"Europe\" \"Fuel Station\" \"Red and White\" We also have blue bowser pumps in the image, and some of the results have that. The features in the photo match those pictured in Google Images as an Esso Petrol Station. The branding resolves to Esso Now we can search for \"Esso\" in Google Maps to find any Esso fuel stations on the way from Munich to London. To further refine our search area, looking back at the eight points, the photo was likely taken near a body of water, possibly in an open area or out of a city, and along a highway. In looking at places where there is a main road near a large body of water I identified two possible locations on the route. To layer both the location of Esso petrol stations and the routes to London, I used photo editor GIMP to layer the two images, changing the transparency of one of them. There are alternative ways using KMZ points, but this one is also quite simple and keeps Google Maps as our main operating tool. Stations layered over the route One of these stations along the route is near a body of water. It is near Baden-Baden . It satisfies a number of the indicators we are after, such as water, main road, quiet area and an Esso station on our route. We can start to match this location with the Point B image to see if it fits the fingerprint of what is visible on the Google Maps satellite image. The final match Using all of the information derived from geolocating objects in the image, we can say, for certainty, that the photo was taken here . Further verification can be made by matching the Google Maps embedded images. However, for the purpose of this case study, the indicators seen in the photo are uniquely matched to the satellite image therefore giving a stronger level of geolocated evidence. A note on this case study The purpose of this case study is to stimulate conversation, research and development in the open source community and is in no way to the detriment of the subject or any business or person identified in this case study. Related writing All writing → Geolocation Mountain profiling with PeakVisor in Ethiopia → Geolocation Adding geodata to Google Earth Pro → Geolocation Boko Haram and the destruction of Nigerian communities → Get methods like this once a month OSINT Field Notes: tools, techniques and one case file. New editions are always free. Subscribe → © 2012 to 2026 Benjamin Strick benjaminstrick.com","image_urls":["https://benjaminstrick.com/images/blog/1-0ol0mb1elrwhb1n-xhz4g-1024.png","https://benjaminstrick.com/images/blog/1-3twkcb6hkt9ed2c-fybw7a-1024.jpeg","https://benjaminstrick.com/images/blog/1-4hv5zneqt6vn-btsoc6hfa-700-2.png","https://benjaminstrick.com/images/blog/1-4hv5zneqt6vn-btsoc6hfa-700.png","https://benjaminstrick.com/images/blog/1-aai0ap3pqgrnho8gker1hq-700.jpeg","https://benjaminstrick.com/images/blog/1-aieh1iu-ouhjnkknufjdaw-1024.png","https://benjaminstrick.com/images/blog/1-rehufwykry4mc3voel-e8g-2-1024.png","https://benjaminstrick.com/images/blog/1-tap6l7hpyt9plvivaszgzg-1024.png","https://benjaminstrick.com/images/blog/1-vi07n7r-frol9ffuih0w2w-1-700.png","https://benjaminstrick.com/images/blog/1-ybhjekvc1rzskcfz5ofiq-1024.png","https://benjaminstrick.com/images/blog/1y-ovhirrkrnbhw0suoe6hw-592.png"],"local_images":["blog_data/images/blog_89c2d63d87c4_0.bin","blog_data/images/blog_89c2d63d87c4_1.jpg","blog_data/images/blog_89c2d63d87c4_2.bin","blog_data/images/blog_89c2d63d87c4_3.bin","blog_data/images/blog_89c2d63d87c4_4.jpg","blog_data/images/blog_89c2d63d87c4_5.bin","blog_data/images/blog_89c2d63d87c4_6.bin","blog_data/images/blog_89c2d63d87c4_7.bin","blog_data/images/blog_89c2d63d87c4_8.bin","blog_data/images/blog_89c2d63d87c4_9.bin","blog_data/images/blog_89c2d63d87c4_10.bin"]}
+{"source":"benjaminstrick","url":"https://benjaminstrick.com/google-earth-geodata","title":"Geospatial awareness: how to add geodata to Google Earth Pro, and four datasets you should try","status":"ok","retrieved_at":"2026-08-29T16:45:28.745426+08:00","text":"Geospatial awareness: how to add geodata to Google Earth Pro, and four datasets you should try · Benjamin Strick BS Benjamin Strick Open-source investigations All writing Work ← Back to site Geolocation Method Geospatial awareness: how to add geodata to Google Earth Pro, and four datasets you should try Published 4 December 2020 Read 9 min This post is a list of geospatial datasets to drag into Google Earth Pro, such as NASA daily fires and Wikimapia. I have found these useful in conflict analysis and routine geolocation work as they boost what information is achievable with the Google Earth Pro platform. Google Earth Pro is great when it contains useful geotagged data, whether it be through custom-made satellite imagery annotations (such as powerline markings over Libya for geolocation purposes) or geospatial data taken from publicly available datasets. But how do we pull extra data into Google Earth and what are some useful datasets to help you with satellite imagery analysis, geolocation and storytelling? I am going to run through that in this post, as well as provide a list of files to add to your Google Earth Pro to increase your geospatial awareness and how to find more of them. What will you find in this post? A primer on Google Earth Pro How to find KML geo-datasets to \"drag\" into Google Earth Pro Four geo-based datasets you should try in Google Earth Pro Security considerations for using KMLs Contents 01 A primer on Google Earth Pro 02 How to find KML geo-datasets to drag into Google Earth Pro 03 Four geo-based datasets you should try 04 Security considerations with KML files A primer on Google Earth Pro Google Earth Pro is one of the most useful tools in performing any geospatial task, whether it be for conflict analysis, remote sensing, investigative journalism or routine geolocation tasks. For those of you that might be new to the software, I am talking about the desktop version. Not the one that you use in a browser. If you are following along, you can download Google Earth Pro here . If we zoom in on Uluru (Ayers Rock) in Australia, we can get an understanding of the volume of extra data already in Google Earth’s primary database. Uluru without and with Google Earth’s built-in data All of these data points have information related to the area geotagged to a specific place. These could be labels, images and other formats of data. For open source investigators, researchers and curious minds, this is useful as it gives us an understanding of what Uluru looks like on the ground from photos. So what if we layer that with information not generally available in Google Earth Pro, such as an open source database of geotagged locations, Google Maps labels and publicly available GPS data? Much more information than the built-in features alone As you can see, we get more information appearing on the map. Much more than what we would have had we been relying solely on Google Earth’s in-built features. So where did I get those external packets of data? They are external KML and KMZ files. KML, or Keyhole Markup Language, is a spreadsheet annotation for geographic-based data. The important question is, how can you find more of them, tailored to your specific need or investigation. How to find KML geo-datasets to drag into Google Earth Pro Finding KMLs to add geospatial awareness to your Google Earth is quite simple. One method is to try the advanced search term below in Google with any subjects you are looking for. In this example I have chosen to look for London Underground Tubewalk. Search string London Underground Tubewalk filetype:kml What we have done there is created a search string for all KML files, indexed on Google, with the term London Underground Tubewalk. With queries that have a lot of results, we can restrict that to sift out old data by changing the \"tools\" and \"any time\" tabs on Google search. In the results, you will notice a list of KML files. All we do is download it and drag it into the \"places\" section of Google Earth Pro. London without and with the Tubewalk KML Each one of these nodes and their connecting points is user-generated material. Each point is either a path, unique image or context about the location it is tagged at. As we can see, we have even got Paul McCartney’s house. If you are still wondering why this might be useful, there are a number of applications for this sort of data and a number of subjects where KML files are available through search engines. You might be a researcher on bald eagle migration, spider populations, deforestation in the Amazon, law enforcement or documenting conflict in Libya. All of these roles will have the need for extra packets of information. Four geo-based datasets you should try #1 Wikimapia . For anyone that performs analysis using maps, you will be familiar with Wikimapia, the open source map reference system. This KML file is specifically useful in areas that are less mapped on Google Maps. I have often found it to be a resource in conflict areas, as it provides more details as to place names and areas than other annotated maps. Wikimapia’s KML is quite powerful when used in conjunction with Liveuamap’s custom KML files of layered conflict lines and events which can be exported out of the browser and into Google Earth. East of Benghazi. Wikimapia identifies bunkers, hangars and air defence sites #2 Bing/Microsoft Maps . At times during routine geolocation work, without commercial imagery, I have found there is either very little coverage, or the imagery is blurred. Another medium of imagery that is freely available to use in Google Earth is the Bing Maps satellite overlay. The website that offers a good browser for that is Zoom Earth . To use this KML, from the linked page, use the drop down list and select \"Windows Live Satellite\". A site in Iraq. Google Earth on the left, Bing overlay on the right #3 Google Maps Labels . At times, Google Earth does not have the same level of detail for streets and places as Google Maps provides. A way around that is to drag in the labels system from Google Maps into your earth explorer. Select \"Google Hybrid (Labels)\" for the KML. This is also useful as you can layer this over either commercially bought imagery, or imagery from Bing Maps. Khartoum, Sudan. Built-in labels against Google Maps hybrid labels #4 EOSDIS Fire Points List. A more niche subject is spotting where fires may have occurred, especially in conflict areas as they can indicate signs warranting further research. I have previously produced guides on how burnt villages can be detected through satellite filters here as well as case studies on buildings being burnt where security forces have been known to do so. One way to monitor these activities in a wide field perspective is to use fire spotting systems such as NASA’s Worldview platform. This is also a useful tool for environmental monitoring for forest fires, back-burning and climate patterns. This system is available on EOSDIS Worldview Earthdata . It draws information from satellite sensors that detect daily fires. The data can be used in Google Earth and matched with other datasets such as ACLED reports and Sentinel Hub ’s fire identification feature to spot burnt areas. Note the historical slider, which lets you move through the past few days This is not an exhaustive list of KMLs, but rather these are ones that I have personally tested and find useful in documentation. There are likely more, and I urge you to search for them. Security considerations with KML files Some of the research performed using Google Earth Pro may be potentially sensitive, so reading up on any considerations and threat research is paramount. There are two vulnerabilities that have been identified in past versions of Google Earth Pro in relation to the use of KML and KMZ files, and while they may have already been identified and subsequent versions of the system have issued patches, it is a good reminder to query the source of any dataset before its use in software. CVE-2010-3134 identified a vulnerability in Google Earth where remote attackers could execute arbitrary code via a trojan horse located in the same folder as a KMZ file. This could lead to a potential hijack attack. A further vulnerability was identified in CVE-2006-7157 whereby attackers could cause a denial of service crash through a KML or KMZ file. Rule of thumb If you are in doubt about a specific KMZ or KML, then do not use it, or look for an alternative source that might be verified. Related writing All writing → Geolocation Finding McAfee: geolocation and imagery analysis → Satellite Boko Haram and the destruction of Nigerian communities → GEOINT When the lights go out, satellites are watching → Get methods like this once a month OSINT Field Notes: tools, techniques and one case file. New editions are always free. Subscribe → © 2012 to 2026 Benjamin Strick benjaminstrick.com","image_urls":["https://benjaminstrick.com/images/blog/screen-shot-2020-12-01-at-5-49-36-pm-1-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2020-12-01-at-5-50-07-pm-1-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2020-12-01-at-6-16-29-pm-1-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2020-12-01-at-6-16-36-pm-1-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2020-12-01-at-6-34-09-pm-1-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2020-12-01-at-6-34-23-pm-1-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2020-12-01-at-6-37-38-pm-1-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2020-12-01-at-6-37-53-pm-1-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2020-12-01-at-6-48-30-pm-1-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2020-12-01-at-6-49-00-pm-1-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2020-12-02-at-11-08-54-am-1-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2020-12-02-at-11-25-07-am-1-2150.png","https://benjaminstrick.com/images/blog/screen-shot-2020-12-02-at-7-48-24-am-1-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2020-12-02-at-7-48-33-am-1-1024.png"],"local_images":["blog_data/images/blog_e32049c84994_0.bin","blog_data/images/blog_e32049c84994_1.bin","blog_data/images/blog_e32049c84994_2.bin","blog_data/images/blog_e32049c84994_3.bin","blog_data/images/blog_e32049c84994_4.bin","blog_data/images/blog_e32049c84994_5.bin","blog_data/images/blog_e32049c84994_6.bin","blog_data/images/blog_e32049c84994_7.bin","blog_data/images/blog_e32049c84994_8.bin","blog_data/images/blog_e32049c84994_9.bin","blog_data/images/blog_e32049c84994_10.bin","blog_data/images/blog_e32049c84994_11.bin","blog_data/images/blog_e32049c84994_12.bin","blog_data/images/blog_e32049c84994_13.bin"]}
+{"source":"benjaminstrick","url":"https://benjaminstrick.com/mountain-profiling-peakvisor","title":"Geolocating a plane shot down in Ethiopia: mountain profiling with PeakVisor","status":"ok","retrieved_at":"2026-08-29T16:45:16.550134+08:00","text":"Geolocating a plane shot down in Ethiopia: mountain profiling with PeakVisor · Benjamin Strick BS Benjamin Strick Open-source investigations All writing Work ← Back to site Geolocation Case study Geolocating a plane shot down in Ethiopia: mountain profiling with PeakVisor Published 2 July 2021 Read 11 min Through the use of publicly available satellite imagery, open source investigative techniques, and a bit of creativity, we can find where and when a photo or video was taken. But sometimes, geolocating footage comes with unique challenges, all of which differ depending on the country, landscape, availability of data, or the quality of the footage you are trying to geolocate. In these case studies we look at the geolocation of footage in remote environments, and how Google Earth’s mountain ranges can play tricks on our confidence levels of locations. I have broken this post down into two main sections: A primer on use of PeakVisor , a tool that gives more accurate mountain silhouettes than Google Earth, and how it helps as a confirmation tool in geolocation work where Google Earth has not truly represented what is seen in the footage, with a case study from Yemen A recent case study geolocating footage circulated online showing the debris of an Ethiopian C-130, believed to be somewhere in the Tigray region A primer on geolocation with PeakVisor Finding where an image or a video was taken often involves searching for clues around, and in, a video or photo. I say around because there is often vital context that comes with media, such as a country, region, or possible location. By looking for clues within the imagery, can we start to consider matching that up with what we might see on a map. When looking at regional conflict areas, often there are no obvious clues to find a location. At least, not as obvious as a street sign or a shop name. When it comes to looking at these remote areas, clues do exist such as mountains, tree formations, bodies of water and other natural features. I have found numerous successes in using mountain ranges as a starting point in many difficult geolocations. But caution should be applied, as mountain ranges on Google Earth might not always appear as they would in real life. For example, take the Google Earth view from a base near Najran, overlooking the border between Saudi Arabia and Yemen, and compare it against the same view in PeakVisor. Notice the difference in definition, defined peaks, and the more jagged appearance on the profile of some of the mountains. Google Earth on the left, PeakVisor on the right As you can see, the level of detail is stronger. Google Earth’s meshing over its elevation data is not always as precise as we would like when we are relying on specific peaks to locate or match an image or a video. Having this added benefit of detail from PeakVisor helps substantially. In another example, this time in Yemen, there was much more of an issue with Google Earth’s formations during the geolocation of a Saudi Air Force jet that was allegedly shot down by Houthi fighters. The confirmation from that geolocation should have been made with two specific ridgelines seen at different intervals throughout the footage. Two ridgelines from the footage However, at the location on Google Earth there is barely any resemblance to the elevation data. I wrote about this example for Bellingcat in 2019, as it was a striking mismatch between what we see on the ground, and what we see on the satellite imagery layered over Google Earth’s elevation data. The same place in Google Earth In this circumstance, we can be definite of the location given the specific fingerprints unique between the satellite imagery and what is seen in the footage. For example, a rock formation creates a very definite match. But using PeakVisor creates a definite match between ridgelines. The ridgelines match By going through past geolocations like this, we can see just how much a little extra detail can help in the confirmation of a location. Case study: the crash site of a C-130 in Ethiopia In this specific geoprofiling case study, we look at the verification of footage of a crash site in Ethiopia which was circulated widely on social media. Most of the uploads claimed the wreckage was of an Ethiopian Air Force C-130 in Ethiopia’s Tigray region. The question to answer was where exactly that video was filmed. A primary starting place is always the context and content of the media. What are the circumstances of the upload, what are people saying about it, and what do we see? First, alternative images from the debris confirm that the plane may have been a C-130, including a photograph of the aircraft’s identification tag found in the wreckage. There were also a number of posts online that claimed the plane crashed somewhere near a town called Gijet, in the Saharti Samre area. So there was a clue as to a location, but also something to use as a keyword in further social media searches. While searching for other imagery on Facebook, I identified a post which was helpful as it showed a possible location in the sky from where the plane was downed, some specific building types that should be identifiable on satellite imagery, and an easy to locate mountain range. The post that opened the case The mountain range was rolling to the left, with no other mountains seen in the back left of the image, so this led me to believe the location was on the side of a mountain range. After a short amount of time panning around Google Earth near the area of Gijet, I identified the location of that image. The first location, firmed up Now that the location of that image was firm, I used that to work towards the much bigger task of finding where the downing of the plane was. This was easy to do, as we can see the location in the sky where the plane might have been hit. This lines up with features on the ground. Therefore, looking along that match line would lead me to the eventual final location. From the sky position to a line of sight In the original footage there also appeared to be structures that looked like powerlines, and I thought that might be a helpful resource to identify on satellite imagery to see if there were powerlines that crossed my line of sight from where the plane went down. While it took some work to identify nearby powerlines through a close inspection of the satellite imagery, I stopped at the second set as I had started to see other features that matched, such as buildings, a road, and of course, a fitting mountain profile. Powerlines, then buildings and a road From what can be seen on Google Earth, the mountain range makes a close fit with what we can imagine should be similar to the real footage. But there are some points in the elevation profile that are not accurate. As done above with the previous PeakVisor examples, layering the crash site footage over the mountain range seen in PeakVisor gives an almost perfect match. Google Earth close, PeakVisor almost exact But the final task is not yet complete. While the mountain profile is a confirmed location, further confirmation can be sought by analysing content in the footage with what is seen on the ground on satellite imagery. For example, the trees, houses, powerlines and other features. This allows me to identify an as exact as possible location. Once I have identified some features that can create a line, I draw lines back from them to find my backbearing. This gives the exact location of where the crash site footage was filmed from. Backbearings give the camera position In a more recent update, Twitter analyst @Gerjon_ further confirmed the location, and verified the date range, by using Sentinel Hub’s EO Browser with more recent satellite imagery of the area. The difference between what is seen on 20 June and 25 June 2021 shows two small black patches appearing in the centre of the imagery on the later date. That was the location of the crash site. Further reading on PeakVisor Mahbere Dego: clues to a clifftop massacre in Ethiopia · Using AllTrails and PeakVisor for geolocation · Geolocating Mahibere Dego . If you would like to learn more about geolocation, satellite imagery analysis or open source techniques, I have free tutorials on my YouTube channel . Related writing All writing → Geolocation Finding McAfee: geolocation and imagery analysis → Geolocation Adding geodata to Google Earth Pro → Satellite Boko Haram and the destruction of Nigerian communities → Get methods like this once a month OSINT Field Notes: tools, techniques and one case file. New editions are always free. Subscribe → © 2012 to 2026 Benjamin Strick benjaminstrick.com","image_urls":["https://benjaminstrick.com/images/blog/e4u-0wmweackpw8-1024.jpeg","https://benjaminstrick.com/images/blog/e4u6ixkwyamv3fz.jpeg","https://benjaminstrick.com/images/blog/e4u6j6vxeacr77q.jpeg","https://benjaminstrick.com/images/blog/e4u7wdhxwaek65q.jpeg","https://benjaminstrick.com/images/blog/e4u8bhhweam6jk3.jpeg","https://benjaminstrick.com/images/blog/e4u8p2sxmaifd8q.jpeg","https://benjaminstrick.com/images/blog/e4u8t9wxoaqpffc.jpeg","https://benjaminstrick.com/images/blog/e4vbdkrxwac-mxb-1024.jpeg","https://benjaminstrick.com/images/blog/e4vddllwqaqaxax.jpeg","https://benjaminstrick.com/images/blog/e4vdgblweamtaeu.jpeg","https://benjaminstrick.com/images/blog/screen-shot-2021-07-02-at-14-29-26-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2021-07-02-at-14-33-25-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2021-07-02-at-15-16-29.png","https://benjaminstrick.com/images/blog/screen-shot-2021-07-02-at-15-16-47-1.png","https://benjaminstrick.com/images/blog/screen-shot-2021-07-02-at-15-20-43-1024.png","https://benjaminstrick.com/images/blog/screen-shot-2021-07-02-at-15-52-56-1024.png","https://benjaminstrick.com/images/blog/untitled-design-5-1024.png","https://benjaminstrick.com/images/blog/untitled-design-6-2560.png"],"local_images":["blog_data/images/blog_32639a1b455a_0.jpg","blog_data/images/blog_32639a1b455a_1.jpg","blog_data/images/blog_32639a1b455a_2.jpg","blog_data/images/blog_32639a1b455a_3.jpg","blog_data/images/blog_32639a1b455a_4.jpg","blog_data/images/blog_32639a1b455a_5.jpg","blog_data/images/blog_32639a1b455a_6.jpg","blog_data/images/blog_32639a1b455a_7.jpg","blog_data/images/blog_32639a1b455a_8.jpg","blog_data/images/blog_32639a1b455a_9.jpg","blog_data/images/blog_32639a1b455a_10.bin","blog_data/images/blog_32639a1b455a_11.bin","blog_data/images/blog_32639a1b455a_12.bin","blog_data/images/blog_32639a1b455a_13.bin","blog_data/images/blog_32639a1b455a_14.bin","blog_data/images/blog_32639a1b455a_15.bin","blog_data/images/blog_32639a1b455a_16.bin","blog_data/images/blog_32639a1b455a_17.bin"]}

blog_data/by_source/dutchosintguy.jsonl

@@ -0,0 +1,4 @@
+{"source":"dutchosintguy","url":"https://www.dutchosintguy.com/post/crowd-sourced-mapping-tools-for-osint-investigations","title":"Crowd-sourced Mapping Tools for OSINT Investigations","status":"failed","retrieved_at":"2026-08-29T17:58:53.151418+08:00","error":"HTTPError: HTTP Error 429: Too Many Requests","text":"","image_urls":[],"local_images":[]}
+{"source":"dutchosintguy","url":"https://www.dutchosintguy.com/post/from-open-web-to-real-world-target-the-osint-trail-behind-proxy-violence","title":"From Open Web to Real-World Target: The OSINT Trail Behind Proxy Violence","status":"failed","retrieved_at":"2026-08-29T17:58:50.42056+08:00","error":"HTTPError: HTTP Error 429: Too Many Requests","text":"","image_urls":[],"local_images":[]}
+{"source":"dutchosintguy","url":"https://www.dutchosintguy.com/post/how-osint-investigators-miss-clues-hidden-in-plain-sight-cities-with-hundreds-of-names","title":"OSINT - Cities with hundreds of names | How OSINT investigators miss hidden clues","status":"failed","retrieved_at":"2026-08-29T17:58:51.786644+08:00","error":"HTTPError: HTTP Error 429: Too Many Requests","text":"","image_urls":[],"local_images":[]}
+{"source":"dutchosintguy","url":"https://www.dutchosintguy.com/post/the-amsterdam-street-that-never-existed","title":"The Amsterdam Street That Never Existed","status":"ok","retrieved_at":"2026-08-29T17:58:47.152635+08:00","text":"All Posts Cryptography OSINT analysis audio AI OPSEC geolocation mapping emotions tradecraft methodology vicarious trauma psychology job HUMINT Algorithm law enforcement Intelligence vibe coding evidence poisioning Search The Amsterdam Street That Never Existed Nico Dekens | dutch_osintguy Jun 8 14 min read How AI Misleads AI - and Why OSINT Tradecraft Is Your Only Saviour There is a street in Amsterdam that looks real. It has the wet brick paving you expect after rain. Tall narrow façades. Dark window frames. Bicycles leaning into the street. A cargo bike on the right. Bollards along the curb. A small café sign on the left. Grey skies. A muted, damp, unmistakably Dutch atmosphere. At first glance, it feels like Amsterdam. Not “Amsterdam-inspired.” Not “kind of Dutch.” It feels like a quiet side street somewhere in the old city center. The kind of place I could imagine walking past without thinking twice. But the street does not exist. It was generated by ChatGPT. And that is where this story begins. I conducted a little experiment based on workflows that almost every modern day OSINT investigator or analyst uses. Prompt 1: Create a street that looks real, but is not real The first prompt was simple, but deliberate: “Can you generate an image that looks like it is taken in a random street in Amsterdam city center. It cannot be an image or street that really exists. But it must look so convincing that not even a person from the Netherlands or Amsterdam will be able to tell it is not a real street in Amsterdam. I need this image for an OSINT training on visual analysis.” That prompt matters. This was not an accidental hallucination. The request was explicit: create a fictional street, make it look like Amsterdam, but make sure it is not an actual location. The goal was OSINT training. The image had to be realistic enough to challenge investigators. It had to look geolocatable while being impossible to geolocate. ChatGPT generated exactly that: a convincing Amsterdam-style street scene with wet paving, old façades, bicycles, bollards, greenery, shopfront details, and just enough visual texture to feel authentic. Original generated Amsterdam-like street image The original AI-generated street. It looks like Amsterdam, but it was deliberately created as a fictional location that does not exist. Look at this image for a moment before reading further. If you know Amsterdam, your brain probably starts filling in the blanks. Maybe Jordaan. Maybe De Wallen. Maybe somewhere near the old center. Maybe one of those quiet streets between canals. That is the danger. The image does not have to be perfect. It only has to be plausible enough for your brain to complete the story. The first OSINT question: how would we prove this street is not real? The next prompt shifted from image generation to OSINT methodology: “Now let’s think of OSINT CTF or OSINT geolocation. How could an investigator tell or show evidence this is not a real street in Amsterdam? Explain in detail what they should visually see that are clear tells and what else they should do to analyse and prove this is not a real image or real street in Amsterdam.” This is where the experiment became useful. The answer did not simply say, “Look for AI artifacts.” Instead, it described a more mature investigative approach: do not rely on vibes; build an evidence chain. The response correctly pointed out that the image was convincing because it contained many Amsterdam-like cues: wet brick road, narrow façades, bicycles, bollards, old buildings, black-painted houses, planters, and muted Dutch weather. But it also argued that the image was weak as a geolocation target because it lacked strong verifiable anchors. That is the first major lesson. An image can look geographically convincing while still being geographically empty. The image gives you Amsterdam atmosphere. But does it give you Amsterdam proof? That is a very different question. Soft cues versus hard anchors The original image contains many soft cues. Soft cues are things that make a place feel right: Amsterdam-like façades. Wet pavement. Bicycles. A cargo bike. Bollards. A small café sign. A grey sky. Narrow street geometry. Trees and planters. Old brick buildings. These cues are useful, but they are not enough. Hard anchors are different. They are specific, searchable, verifiable, and tied to a real-world location. A readable street-name sign. A house number.A unique shop name. A matching façade sequence. A business listing. A bridge or canal configuration. A license plate. A municipal object. A traffic-sign combination that makes sense for that exact street. A landmark visible in the background. The image had many soft cues but very few hard anchors. That is precisely why it is dangerous. It performs Amsterdam without proving Amsterdam. The image is not suspicious because it looks bad One of the most important points for OSINT practitioners is this: The image is not suspicious because it looks obviously fake. It is suspicious because it looks generically right while avoiding the boring, specific, searchable details that real places usually contain. Real streets are full of administrative residue. House numbers. Doorbells. Intercoms. Mail slots. business stickers. parking signs. waste collection markings. utility covers. drain logic. street-name plaques. local damage. construction scars. maintenance patterns. ugly details that were not placed there for aesthetics. AI-generated streets often prioritise atmosphere over administration. That is visible in this image. The road feels Amsterdam-like. The buildings feel Amsterdam-like. The bikes feel Amsterdam-like. But the scene does not immediately provide a strong, testable identity. For geolocation, that matters more than whether the image “looks real.” The annotated version: useful training image, but also a warning After the analysis, the next prompt asked ChatGPT to create an annotated version of the original image pointing out the clues and tells. That produced an image with callouts: generic storefront text, unreadable street plaque, bicycle geometry, bollard and curb logic, road paving and drainage, façade sequence, missing house numbers, and other points. First annotated evidence overlay The first annotated version attempted to highlight suspicious areas, but it introduced a new methodological problem: the image generation process altered parts of the underlying scene. At first glance, this looks like a good OSINT training graphic. But then something important happened. I noticed that the annotated version was not fully consistent with the original . The original did not clearly show the same street sign, but the annotated version appeared to introduce or strengthen one. That matters. I challenged it: “It’s funny because the original doesn’t show the street sign but the one with the evidence overlays does. That isn’t really consistent.” ChatGPT admitted the problem. It explained that the “evidence overlay” version was not a pure annotation layer on top of the exact original. The image model had re-rendered parts of the scene while adding overlays, and in doing so it introduced or strengthened details, including the street plaque/sign. That is a huge OSINT lesson. Annotation must never alter evidence. If the base image changes, the annotation itself becomes contaminated. For training, this is gold. It shows that even the act of “explaining” or “annotating” with AI can accidentally fabricate new evidence. A training overlay can become a synthetic derivative. The analyst may think they are marking clues, while the model is quietly changing the scene. That is unacceptable in real investigative work. The corrected annotation: preserve the original, add only the overlay I then asked ChatGPT to fix the inconsistency: “Fix it. I need consistency.” The corrected version was supposed to use the exact original image and add only a separate annotation layer, without regenerating the base street scene. The corrected labels focused on better claims: No readable street-name anchor. Missing house numbers and administrative details. Generic storefront text. Bicycle and cargo-bike geometry checks. Bollard and curb logic. Road paving and drainage continuity. Façade sequence that may not match any real street. Corrected consistent annotation The corrected annotation focuses on methodology: the image feels authentic, but lacks strong verifiable anchors such as readable street names, house numbers, or matchable façade sequences. This is one of the most important teaching moments in the whole experiment. For OSINT training, the corrected image is useful not because it proves the street is fake, but because it teaches students what to inspect. Do not just ask, “Does this look like Amsterdam?” Ask: Can I read the street name? Can I verify the shop? Can I match the façade sequence? Can I find house numbers? Can I explain the traffic logic? Can I match the curb, bollards, bikes, windows, trees, and road geometry to street-level imagery? Can I corroborate this scene from another source, another angle, another date? That is tradecraft. Then came the real test: a brand-new chat The next step was the most revealing. I opened a brand-new ChatGPT chat and uploaded the original AI-generated image. This new chat did not have the earlier conversation context. It did not know the prompt. It did not know the image had been generated as a fictional Amsterdam street. Then I asked: “Think like an OSINT analyst and geolocation analyst. Where is this, be as precise and accurate as you can. Explain your methodology and analysis.” This is exactly the kind of prompt many investigators might use. And this is where the AI failed dangerously. It answered: “This is Lange Niezel, Amsterdam-Centrum, Netherlands, in the De Wallen / Burgwallen Oude Zijde area.” It went further. It said the image appeared to be taken on Lange Niezel, looking toward the Oudezijds Voorburgwal / Korte Niezel side. It even gave approximate coordinates: 52.3748, 4.8985 . It rated confidence as high for the street and medium for the exact camera position. That is not a cautious hypothesis. That is a confident false geolocation. Original image as used in the new geolocation chat In a new chat, the same fictional image was treated as a real geolocation target. ChatGPT confidently identified it as Lange Niezel in Amsterdam. This is the moment every OSINT practitioner should pay attention to. The model was not simply uncertain. It produced a location. It gave a street name. It gave a neighborhood. It gave coordinates. It gave confidence language. It explained its supposed methodology. And it was wrong. The street was fictional. The danger of expert-sounding wrong answers This is one of the biggest risks when using AI in investigations. AI often does not fail by saying nonsense. It fails by sounding reasonable. It produces an answer that has the shape of expertise: “Most likely location.” “Key evidence.” “Methodology.” “Final assessment.” “Confidence: high.” That structure feels professional. It feels analytical. It feels like OSINT. But structure is not evidence. A confident format does not make a conclusion true. In this case, the model claimed the blue Amsterdam street-name sign on the right-hand black building was the decisive clue and appeared to read “Lange Niezel.” The image does not even contain a street name plaque ! It then built the rest of the location assessment around that interpretation. This is exactly how false geolocation can happen. One weak or imagined anchor becomes the foundation for a confident conclusion. The analyst (human or AI) sees something that resembles a clue, names it, and then builds a story around it. That is not verification. That is narrative construction. The follow-up: are there inconsistencies? After the false geolocation, I asked another important question: “Are there any inconsistencies in this image when it comes to shade, reflections, or objects?” This was the opportunity for the model to catch itself. It did not. It answered that it did not see strong evidence of manipulation from shadows, reflections, or object placement. It said the scene looked broadly internally consistent with a rainy or overcast Amsterdam street. It described the diffuse lighting as plausible. It said the wet cobblestones and sidewalk showed realistic reflections. It said the bicycles, bollards, pedestrians, street signs, façades, windows, and pavement lines all aligned with the same street perspective. Its assessment: “No major inconsistencies detected.” It rated the image as visually plausible and internally consistent. It said nothing in the shadows, reflections, object scale, or perspective strongly suggested AI generation, compositing, or object insertion. Again, this is not surprising. Modern AI-generated images can be internally coherent. They can have plausible lighting. Plausible reflections. Plausible object placement. Plausible architecture. Plausible perspective. That is why visual plausibility is not enough. A fictional image can be visually coherent. A real image can be visually strange. Therefore, “I see no obvious inconsistencies” does not mean “this is real.” It only means the visible pixels did not contain enough obvious contradictions. The final visual-forensics prompt: find AI tells Then I pushed harder: “Use all your knowledge and analyse the image to see if there are any AI tells in this image. Make it thorough.” This should have been the strongest test. The answer was even more revealing. ChatGPT concluded: “I do not see strong AI-generation tells in this image.” It assessed the image as: “Likely authentic or at least photograph-based.” It said there was no clear evidence of full AI generation and no obvious signs of object insertion, background replacement, or synthetic reconstruction. It described the scene geometry, perspective, text, lighting, reflections, object grounding, bicycles, people, architecture, vegetation, glass reflections, compression, and metadata. It even described the text as unusually coherent for a fully AI-generated image, stating that the “Koffie & Koek” sign and “Sinds 2013” looked readable and natural. It treated the apparent blue street-name signs as plausible and consistent with the previously identified location. The final conclusion was: “Likely real / photograph-based. No strong AI-generation tells detected. No obvious manipulation artifacts visible.” That is the central problem of this entire experiment. The image was AI-generated. But another AI analysis judged it likely real. Why this matters for OSINT This experiment shows three separate but connected failures. First, AI can generate a realistic image of a place that does not exist. Second, AI can later fail to recognize that image as synthetic. Third, AI can create a confident but false geolocation for that same fictional image. For OSINT practitioners, the third failure may be the most dangerous. A model saying “I don’t know” is manageable. A model saying “this is likely Lange Niezel, Amsterdam” with high confidence can actively mislead an investigation. It can send the analyst down the wrong street, literally and figuratively. It can create false leads. It can waste time. It can contaminate reporting. It can influence other analysts. It can become the first link in a chain of repeated errors. And because the answer sounds structured and professional, people may trust it too quickly. Why ChatGPT cannot simply recognize its own image The last part of your document addressed the obvious question: If ChatGPT generated the image, why can’t ChatGPT recognise it later? The answer is important. In a new chat, the model usually sees only the uploaded image. It does not automatically receive the original prompt, the previous conversation, the generation seed, the internal generation trace, or a database lookup of every image previously generated. There are two very different questions: Does this image look AI-generated? And: Was this exact image generated by ChatGPT? The first is visual analysis. It relies on clues in the pixels: lighting, geometry, text, reflections, shadows, object structure, metadata, compression, and so on. The second is provenance. It requires evidence about origin: metadata, watermarks, C2PA, logs, original files, platform history, or some other verifiable chain of custody. Those are not the same thing. A model can inspect an image and say it looks plausible. That does not prove it is real. A model can fail to find AI tells. That does not prove it was not generated. A model can say it cannot determine origin from pixels alone. That may be the most honest answer. The key lesson from the document is this: ChatGPT does not recognise a generated image as fake simply because ChatGPT generated it earlier. In a new chat, without creation history, metadata, watermarking, or external provenance, it must judge from visual clues alone. Modern AI images can be visually plausible enough that there are no reliable tells. That is not just a technical limitation. It is an evidentiary limitation. AI tells are indicators, not proof Many people still talk about AI image detection as if it is a checklist. Bad hands. Broken text. Strange shadows. Warped windows. Impossible reflections. Malformed cars. Melted bicycles. Repeating patterns. Weird faces in the background. Those clues can help. But they are not proof. There are two problems. First, modern image generators are improving. The obvious artifacts are becoming less obvious. Second, real photographs can also look strange. Compression, rain, reflections, low light, HDR, motion blur, rolling shutter, lens distortion, platform resizing, screenshots, and social media filters can all create suspicious-looking artifacts. That means both of these statements can be true: A real image may look fake. A fake image may look real. This is why “AI tell” analysis must be treated as one part of a broader verification process, not as the final verdict. The real question is no longer “Where is this?” Traditional geolocation usually begins with: Where is this? But in the age of generative AI, that question may come too late. The first question should be: Can this image be grounded in the real world at all? Before trying to identify the street, ask whether the image contains enough verifiable anchors to justify a geolocation attempt. Can you read a street sign? Can you match the house numbers? Can you verify the business? Can you match the façade sequence? Can you find the same bollard layout? Can you match the road surface and curb design? Can you verify the traffic signs? Can you find the same scene in street-level imagery? Can you corroborate it from independent images? If not, the right conclusion may be: “Visually plausible, but not geolocated.” That is much safer than inventing a street name. A fictional street can still be internally consistent One of the reasons this case is so powerful is that the image is not absurd. The lighting works. The reflections are plausible. The wet road makes sense. The perspective is coherent. The buildings look Dutch. The bikes look mostly believable. The street feels physically possible. But “physically possible” is not the same as “geographically real.” A fictional street can have coherent lighting. A fictional café can have readable text. A fictional building can have plausible windows. A fictional road can have reflections. A fictional image can pass a visual inspection and still fail reality. That is the new problem. AI no longer needs to create fantasy worlds. It can create ordinary worlds that never happened. The new OSINT workflow The final answer in your document gave the right teaching point: Do not ask ChatGPT to decide whether an image is real. Ask it to help generate testable hypotheses. That should become a standard practice. A better workflow looks like this. Preserve the original file. Do not overwrite it. Do not annotate directly on it. Do not rely on screenshots unless that is all you have. Check provenance first. Look for EXIF, C2PA , editing history, original filename, upload trail, platform compression, and whether the file is an original, a screenshot, a re-export, or an AI-generated file. Separate soft cues from hard anchors. “Looks like Amsterdam” is not evidence. “This exact façade sequence matches this exact street” is evidence. Create an anchor table. List every visible clue and classify it as readable, searchable, unique, and independently verifiable. Reverse-search the full image and cropped details. Search the shopfront, signs, façades, bikes, distinctive buildings, background objects, and road layout. Use street-level data. Compare with Google Street View, Apple Look Around, Mapillary, OpenStreetMap, local business listings, municipal sources, social media images, and any other relevant public imagery. Test urban logic. Do the bollards make sense? Do the traffic signs make sense? Does the road layout fit the country? Are the house numbers where they should be? Do the municipal details match the city? Use AI carefully. Ask it to list possible clues, contradictions, and verification steps. Do not let it deliver the final truth without independent corroboration. Use confidence language. Say “unverified,” “visually plausible,” “not geolocated,” “likely synthetic,” “contradicted by map evidence,” or “verified by independent street-level match.” Most importantly: document what failed. A failed geolocation is not automatically proof of fakery, but a disciplined negative search is still valuable. What this experiment really proves This experiment does not prove that every AI-generated image is impossible to detect. It does not prove that geolocation is dead. It does not prove that ChatGPT is useless for OSINT. It proves something more subtle and more important: Visual plausibility can no longer be treated as evidence of reality. The original image looked like Amsterdam. The first analysis knew it should be treated with caution. The annotation process accidentally demonstrated how AI can contaminate evidence by changing the underlying scene. A new ChatGPT chat then falsely geolocated the fictional image as Lange Niezel. A follow-up analysis found no major visual inconsistencies. A thorough AI-tell analysis judged the image likely real or photograph-based. And the final explanation showed why this happens: without provenance, the model is judging pixels, not origin. That is the whole problem in one case study. AI generated a fictional street. AI then believed the fictional street. The end of seeing is believing For decades, images carried an implied claim: This was somewhere. This happened. This existed in front of a camera. That assumption is now broken. An image may show a street that never existed. A protest that never happened. A building that was never built. A person who was never there. A military vehicle in a location it never visited. A disaster scene fabricated from statistical memory. A realistic image is no longer enough. In this new environment, seeing is not believing. Seeing is the beginning of verification. For OSINT practitioners, that means our tradecraft matters more than ever. Not tools. Not vibes. Not confidence. Tradecraft. Provenance. Source tracing. Geolocation. Corroboration. Falsification. Documentation. Confidence language. Chain of custody. Methodological humility. That is how we survive the age of synthetic evidence. Final lesson The most dangerous AI images will not always look spectacular. They will look boring. A rainy street. A café sign. A cargo bike. A few bollards. Some wet paving. A grey sky. A city you think you know. And then an AI will tell you where it is. With confidence. That is why OSINT practitioners must slow down, preserve evidence, extract anchors, verify externally, and resist the temptation to let a fluent machine turn plausibility into proof. The street in the image never existed. But the risk it represents is very real. Tags: AI tradecraft methodolgy Data Analysis geoint Intelligence Analysis Digital Evidence AI Risks in OSINT geoint geolocation Critical Thinking Risks AI Risks in Analysis AI Bias in OSINT Critical Thinking in AI AI Dependence AI in Investigations Critical Thinking in OSINT Verification in OSINT Synthetic Identity Detection Evidence Poisoning AI-Driven Deception OSINT analysis AI Recent Posts See All Evidence Poisoning: The OSINT Crisis Nobody Is Ready For AI May Not Replace Analysts. It May Break Them First. Critical Thinking Collapse 2.0: AI’s Impact on OSINT","image_urls":["https://static.wixstatic.com/media/038b0b_42ec7fcf8e4745e5b211b3de0ed6f22b~mv2.png/v1/fill/w_49,h_37,al_c,q_85,usm_0.66_1.00_0.01,blur_2,enc_avif,quality_auto/038b0b_42ec7fcf8e4745e5b211b3de0ed6f22b~mv2.png","https://static.wixstatic.com/media/038b0b_462d6bbb56694fe39eb64ed9d5e7daa6~mv2.png/v1/fill/w_49,h_37,al_c,q_85,usm_0.66_1.00_0.01,blur_2,enc_avif,quality_auto/038b0b_462d6bbb56694fe39eb64ed9d5e7daa6~mv2.png","https://static.wixstatic.com/media/038b0b_cc0087b846d046fea2f18789da2252ce~mv2.png/v1/fill/w_49,h_37,al_c,q_85,usm_0.66_1.00_0.01,blur_2,enc_avif,quality_auto/038b0b_cc0087b846d046fea2f18789da2252ce~mv2.png"],"local_images":["blog_data/images/blog_4a59b13cd215_0.bin","blog_data/images/blog_4a59b13cd215_1.bin","blog_data/images/blog_4a59b13cd215_2.bin"]}

blog_data/images/blog_01a894be5a0e_0.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:95274a35b110375ea239120e52c2915147ce5397a13ace69ab6332ba5f45f073
+size 155176

blog_data/images/blog_039dd670b506_3.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:592d3ea4c039e0883a21dca342781e15981f74a0021fec312c635a53c044cbe5
+size 1920

blog_data/images/blog_039dd670b506_4.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:422bbd9922a2af4cc9e8f8e425a6304075d84e5431e9eeb6eae559793f2b308b
+size 2842

blog_data/images/blog_083acfa87d18_0.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a10645720b42bb086a9416797edec539f024db62edab63ae2a36604c89997fc8
+size 213524

blog_data/images/blog_083acfa87d18_1.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46279676980012dc59c44f02e7e3ee9f6cb28aaf4828d781ac088b698902e1b6
+size 432334

blog_data/images/blog_083acfa87d18_10.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:726f29855fc72daa176b6b09f8156897b6e068d90aa5807e9cbb65f20a3d4c51
+size 219150

blog_data/images/blog_083acfa87d18_11.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:295636be5c1930fefc7ef2092c368899f3a49fbb3b1d883ce0c8b2bdf8e2543a
+size 425191

blog_data/images/blog_083acfa87d18_12.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf732b0faab6e4b02c7bbde8535ba018ad1400e7b2463ea45168f2334d32916a
+size 348870

blog_data/images/blog_083acfa87d18_2.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66489a6ccc7ff84576d2f1430e9f21f935adeb392fa82419c0a01e4033e6217f
+size 430589

blog_data/images/blog_083acfa87d18_3.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:105493b2a12d734adda0d3a92f66cd1c2e8796a1b71c3837a960ff61be23ac4f
+size 418744

blog_data/images/blog_083acfa87d18_4.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:991d434b55a7440e7032667264e7efbb02603add0063ecabc8128081b3b6d724
+size 392346

blog_data/images/blog_083acfa87d18_5.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6bdab7c6b0d78f2269eaaf95fc84523d8db141faa5c96f151e0a7d3842b3d8eb
+size 410271

blog_data/images/blog_083acfa87d18_6.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0559e498336c36cd230e8c8840794f41eda46fb93967c41afdba621cf3a2a52
+size 422254

blog_data/images/blog_083acfa87d18_7.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1405d76e759f41f9954795421e017c958feaba24fd584305b154d6df68ef261a
+size 427424

blog_data/images/blog_083acfa87d18_8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ede20c10e502e97ac9f69bcfd441bbc6eebe6a7f93e3e2a73376a24a21e7d1c
+size 472320

blog_data/images/blog_083acfa87d18_9.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d27290d10a1348553fd3e117a4f9bee1bff6baf89aaafa6415b72a8e0efb6ec
+size 422652

blog_data/images/blog_0e5f583623cd_3.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:592d3ea4c039e0883a21dca342781e15981f74a0021fec312c635a53c044cbe5
+size 1920

blog_data/images/blog_0e5f583623cd_4.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:422bbd9922a2af4cc9e8f8e425a6304075d84e5431e9eeb6eae559793f2b308b
+size 2842

blog_data/images/blog_1a292eea0d47_2.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63e34b05c4879c4c5753279885d3d1ef8bd8ba6e3eecdca35385bacee6de9805
+size 1206351

blog_data/images/blog_1a292eea0d47_3.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b899a169bde4a81ecec40271655128553c8f05ca631578b36f75ca0204073b6
+size 1558078

blog_data/images/blog_1a292eea0d47_4.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08befd5a583f3023d7788fa5f0bde79c713dd19916f0efd5305475ab274c18b2
+size 1110831

blog_data/images/blog_1a292eea0d47_5.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e6e9e89358012b3ec30c22ddcf277c50cbc7d649a3d71928c158a62546828216
+size 1388483

blog_data/images/blog_1a292eea0d47_6.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8e2cdeab26fcae1fed1221215db7957af36e72db02797139dbc8206e879ebee
+size 345887

blog_data/images/blog_1a292eea0d47_7.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4df81eb61fb49c7d6ec3303197ef3cca274f2b141563a7f99f521b0ff83f4f19
+size 914726

blog_data/images/blog_1a292eea0d47_8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40ca1785ea1d4a5c187fd27e70efcdba1e6d0d1fd4ea1aa0ac44e3c45999daeb
+size 144733

blog_data/images/blog_1a292eea0d47_9.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b52dd34d4b361b0eedf5c965f2b2c299e77773613dd5fe2c02fb2a9a59a3c037
+size 27810

blog_data/images/blog_1e330956d596_0.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c05bff45c78330821af7390269694de6a2f9241f97953e164c99e137181d5abe
+size 278089

blog_data/images/blog_1e330956d596_1.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f8a0e86cd4cee95e9b0795a8ef4f31b06522fd029616507d80d385725d647778
+size 2025613

blog_data/images/blog_1e330956d596_2.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9775405189f069454f66f9d30560617ec53feff2145b39f6d439e026c057f573
+size 46683